In H.264/AVC, one of the standard specifications for image encoding schemes, it is possible to use different quantization steps for each component of the orthogonal transform coefficients when quantizing image data in the High Profile or higher profile. A quantization step for each component of the orthogonal transform coefficients may be set on the basis of a quantization matrix (also called a scaling list) defined at the same size as the units of orthogonal transform, and a standard step value.
FIG. 28 illustrates four classes of default quantization matrices which are predefined in H.264/AVC. The matrix SL1 is the default 4×4 quantization matrix for intra prediction mode. The matrix SL2 is the default 4×4 quantization matrix for inter prediction mode. The matrix SL3 is the default 8×8 quantization matrix for intra prediction mode. The matrix SL4 is the default 8×8 quantization matrix for inter prediction mode. The user may also define one's own quantization matrix that differs from the default matrices illustrated in FIG. 28 in the sequence parameter set or the picture parameter set. Note that in the case where no quantization matrix is specified, a flat quantization matrix having an equal quantization step for all components may be used.
In High Efficiency Video Coding (HEVC), whose standardization is being advanced as a next-generation image encoding scheme to succeed H.264/AVC, there is introduced the concept of a coding unit (CU), which corresponds to a macroblock of the past (see Non-Patent Literature 1 below). Furthermore, one coding unit may be split into one or more prediction units (PUs), which signify the units of a prediction process. Intra prediction or inter prediction is then conducted on each prediction unit. In addition, one coding unit may be split into one or more transform units (TUs), which signify the units of orthogonal transform. Each transform unit is then subjected to an orthogonal transform from image data into transform coefficient data, and the transform coefficient data is quantized. Non-Patent Literature 2 below proposes a technique called the short distance intra prediction method, which enables the selection of comparatively small-size, non-square prediction units (line or rectangle prediction units, for example) in intra prediction modes. In this case, the shapes of the transform units may also become non-square to match the shapes of the prediction units.